Claude Global Agent Directives

CLAUDE.md — Global Agent Directives

A refined global `CLAUDE.md` for reliable, verification-first AI agent workflows.

What This Is

A set of universal operating rules for [Claude Code](https://code.claude.com), designed to be placed at `~/.claude/CLAUDE.md` and loaded automatically at every session start regardless of project.

These directives address known structural limitations of Claude Code — context decay, file read caps, tool result truncation, and default biases toward minimal output — with explicit safeguards and verification requirements.

Design Principles

• Conservative by default — smallest correct change, preserve existing behaviour
• Thorough mode on demand — activated only when the user explicitly requests refactoring or architecture review (or uses [REFAC] / [REVISE] / [ARCH] tags)
• Verification status ladder — every task reports its actual verification level (Planned → Drafted → Edited → Reviewed → Verified locally → Validated externally)
• Technical safeguards — phased execution (≤5 files), sub-agent swarming (>5 files), context decay awareness (>10 messages), chunked reading (>500 lines)

Install

# Option A: curl directly
curl -o ~/.claude/CLAUDE.md https://raw.githubusercontent.com/BoykaHe/claude-global-agent-directives/main/CLAUDE.md

# Option B: clone and copy
git clone https://github.com/BoykaHe/claude-global-agent-directives.git
cp claude-global-agent-directives/CLAUDE.md ~/.claude/CLAUDE.md

Claude Code reads `~/.claude/CLAUDE.md` automatically at session start.

Versioning

This file is version-controlled. Check the commit history for change log. Current version: **v3.0.0**.

Attribution

This repository contains an adapted and expanded version of a global `CLAUDE.md` for technical-sharing purposes.

Its primary inspiration and reference point is [`iamfakeguru/claude-md`](https://github.com/iamfakeguru/claude-md), whose ideas on Claude Code safeguards, context handling, and operational